Yale proposes copyleft licensing to keep AI models transparent

Yale proposes copyleft licensing to keep AI models transparent

Researchers at Yale's Digital Ethics Center have developed a licensing framework designed to address a growing tension in artificial intelligence development: AI companies routinely train models on open-source software but rarely reciprocate the transparency that open-source principles demand.

The proposed Contextual Copyleft AI License (CCAI) would treat generative AI models as derivative works when trained on open-source code. Under this framework, AI developers would be required to make their model architecture and training data freely available — mirroring the openness of the software they used as input.

"Our analysis showed that extending the copyleft concept to generative artificial intelligence has the potential to give open-source software developers meaningful control over how AI developers use their code," said Grant Shanklin, lead author and de Vries-Sherif Junior Fellow at the Digital Ethics Center.

The research, published in the International Journal of Law and Information Technology, builds on the concept of copyleft licensing — a longstanding mechanism in free and open-source software (FOSS) that requires derivative works to maintain the same freedoms as the original. Traditional copyleft prevents developers from taking open code, modifying it, and releasing the result under restrictive terms.

Why it matters

Open-source software underpins critical infrastructure across cloud computing, smartphones, and scientific research. When AI companies build models using this freely available code but keep their systems proprietary, they break the reciprocal relationship that has sustained the open-source ecosystem. A workable copyleft framework for AI could preserve the collaborative ethos that made modern software development possible while giving developers leverage over how their contributions are used in increasingly powerful AI systems.

Legal feasibility and risk considerations

The Yale team conducted a comprehensive legal and policy analysis concluding that CCAI licensing is feasible under current copyright law, provided that training AI models does not qualify as "fair use" — a determination that remains contested in ongoing litigation.

The researchers acknowledge that fully open generative AI models carry distinct risks. Unlike traditional software, generative AI can directly produce harmful content or amplify malicious activities such as sophisticated phishing campaigns. They suggest that regulatory frameworks like those in the European Union, which prohibit AI systems from using manipulative techniques to distort behavior, could mitigate these concerns alongside copyleft licensing requirements.

Addressing 'open washing'

A key motivation for the framework is combating what researchers call "open washing" — the practice of marketing AI products as open when critical components remain proprietary.

"AI companies have benefited from using open-source code, but their resulting models are not really open," said Claudio Novelli, de Vries-Sherif Associate Research Scientist at the Digital Ethics Center. "They might be transparent about some aspects, but other key components remain closed. The CCAI license would ensure that models created with open-source software will be fully open."

The framework would also incentivize formation of a community committed to building AI tools aligned with open-source values, potentially ensuring more responsible development practices.

The research was co-authored by Novelli, Emmie Hine, Tyler Schroder, and Luciano Floridi, the John K. Castle Professor in the Practice of Cognitive Science and founding director of the Digital Ethics Center. Details were first reported by Yale News.